How to model clitics in the General Linearization Model
Speaker: Ciprian Gerstenberger
Institution: Saarland University
Abstract:
In my previous IGK talks, I proposed a model for linearization based on the mereological concepts of part and whole: the General Linearization Model (GLM). The core part of the GML, the Linear Order Part (LOP), represents a generalization of the traditional concept of the constituent, yet differs considerably from the Word Order Domains proposed by (Reape 1994), (Kathol 2000), based on HPSG, from the linearization approaches taken by (Richter&Sailer 1995), (Penn 1997) based in HPSG, as well, or from the Topological Hierarchies approaches based on (more or less intensive Melc'uk flavoured) Dependency Grammar (DG) by (Gerdes&Kahane 2001), (Bohnet 2004). While these models either use the notion of constituency and/or topological field as a further constraint for the linearization of words, as it is the case with the DG-based approaches, or add further complexity to the already very heavy description apparatus in order to loosen the strictness imposed by a purely constituency-based framework, as it is the case with the HPSG-based approaches, the GLM assumes the full flexibility of any part of an utterance which can be linearized, even that of a phoneme. The idea of GLM originated from the awareness that the smallest linearizable entity in the language is the phoneme, on the one hand, and that we, as linguists, work with only a fairly vague notion of what is a word.
As already mentioned, GLM proposes a single entity type for linearization, the Linear Order Part (LOP), which can be a phoneme, a group of phonemes (not necessarily a syllable), a syllable (not necessarily a morpheme), a morpheme, a word, a constituent phrase, a group of constituents, etc. From a cross-linguistic perspective -- but even within a certain language --, it is hard to tell what precisely is a word and what not. At the boundary between phonology and morphology, on the one hand, and between morphology and syntax, on the other hand, clitics are the best illustration of this issue.
First, I will roughly describe the basic difficulties posed by linguistic entities generally called "clitics": are they affixes, clitics proper or words? Then, based on the phenomenon coined by linguists as "floating affixes", a contradiction in terms (one could say, a floating affix is a married bachelor: something which is an affix, hence, fastened -- from the Latin ad- + figere "to fasten" --, can not float), I will briefly present two HPSG-based descriptions of this phenomenon and their weak points, namely (Kupsc&Tseng 2005) and (Crysmann 2006). Finally, I will show the advantages of GLM for coping with this problem from the point of view of linearization, using a linearization grammar for a small Polish example.